1. L. Young, Synthesis Imaging Summer School, 19 June 2002 1 Spectral Line I Lisa Young This lecture: things you need to think about before you observe After you obtain your data: see lecture by J. Hibbard, Spectral Line II (or A. Peck, spectral line VLBI.)

2. L. Young, Synthesis Imaging Summer School, 19 June 2002 2 What are the major differences between spectral line and continuum observing? Bandpass calibration (Why? How?) Velocity specifications and Doppler Tracking Correlator setups

3. L. Young, Synthesis Imaging Summer School, 19 June 2002 3 Bandpass Calibration: Why? Insert figure showing imperfect spectral response Response to a point source of unit amplitude at the phase center

4. L. Young, Synthesis Imaging Summer School, 19 June 2002 4 Example: what happens if you don’t do bandpass calibration? Insert figure: from my data, pt src uncalibrated.

5. L. Young, Synthesis Imaging Summer School, 19 June 2002 5 Bandpass Calibration: Why? We don’t want to confuse a gain variation on a continuum source with spectral line emission or absorption. Therefore, the quality of the final image will be limited by (among other things) how well the gain variations can be calibrated. If your spectral line is spatially coincident with a continuum source which is 10 times as strong, and gain variations are 10%, your line may be swamped by those gain variations.

6. L. Young, Synthesis Imaging Summer School, 19 June 2002 6 Bandpass Calibration: How? Some instruments will do a rough bandpass calibration for you automatically, e.g. by looking at a flat-spectrum noise source. If your project is not demanding, this may be good enough. (N.B. the VLA is not one of these instruments.) Otherwise, observe a suitable flat-spectrum object. should be bright doesn’t have to be a point source (though you don’t want one which is so large that it is resolved out on many baselines). Use these data and the known spectrum to determine B ij  the gain as a function of frequency, which is then applied to the data. Details are in the Spectral Line II lecture.

7. L. Young, Synthesis Imaging Summer School, 19 June 2002 7 How long should I observe my bandpass calibrator? Errors in B ij (  will contribute to errors in your final image. Thus, want (S/N) BPcal > (S/N) continuum in image. For (S/N) BPcal = 3 (S/N) cont this works out to be t BPcal = t source * 9 (S cont /S BPcal ) 2 Or long enough to get good signal-to-noise in each channel of your bandpass calibrator. Adjust the criterion above as necessary for your science! Normally you are advised to observe the bandpass calibrator twice per run, for redundancy and to help correct for time variability in bandpass shapes (details to follow).

8. L. Young, Synthesis Imaging Summer School, 19 June 2002 8 How long should I observe my bandpass calibrator? (continued) Bandpass calibrator should be observed at same frequencies as target source(s). If you have multiple targets with widely differing frequencies (> few MHz, for VLA), you’ll need multiple observations of the bandpass calibrator. Standard Lore: these simple techniques will allow you to reach spectral dynamic ranges (Peak of strongest continuum in image / rms noise in spectrum) of perhaps 500:1 for VLA data.

9. L. Young, Synthesis Imaging Summer School, 19 June 2002 9 What if I need very high spectral dynamic range? Beware: VLA’s bandpasses are time-dependent! (e.g. Carilli 1991)

10. L. Young, Synthesis Imaging Summer School, 19 June 2002 10 Correcting for time-dependent bandpass shapes Frequent bandpass calibration (30 min. or less), interpolating in time between bandpass observations, and removing the most variable antennas may allow you to reach spectral dynamic ranges of a few thousand to one (at the VLA).

11. L. Young, Synthesis Imaging Summer School, 19 June 2002 11 Bandpass calibration for Galactic HI observations VLA’s C and D configurations are sensitive to Galactic HI emission. All configurations may detect HI in absorption in front of continuum sources. Don’t want a spectral line in your bandpass calibrator! If your target line is at velocities close to zero, you may want to observe the bandpass calibrator at two frequencies offset by about 2 MHz from your target frequency. Average or interpolate later. Standard lore: this technique will limit your spectral dynamic range to 200 or so.

12. L. Young, Synthesis Imaging Summer School, 19 June 2002 12 Example: bandpasses at two different frequencies.

13. L. Young, Synthesis Imaging Summer School, 19 June 2002 13 Velocity Specifications Most commonly, observers will specify the radial velocity of the target source. The telescope will calculate which frequencies are desired. Is your source velocity measured in a heliocentric frame (corrected for the Earth’s motion around the sun) or with respect to the Local Standard of Rest (LSR)? The difference can be up to 20 km/s or so.

14. L. Young, Synthesis Imaging Summer School, 19 June 2002 14 Velocity Specifications (continued) Are you using the radio definition of velocity or the optical definition? (Both are approximations to the relativistic Doppler shift formula.) v radio /c = ( rest  obs  rest  v opt /c =  rest  obs  obs At large velocities, the difference between the two definitions can cause you to miss your line.

15. L. Young, Synthesis Imaging Summer School, 19 June 2002 15 Doppler Tracking: What is it? We are riding on the Earth, which is rotating and moving around the sun. Our velocity with respect to astronomical objects is not constant in time or in direction. Thus, the translation between source velocity and observed frequency is always changing. In Doppler tracking, the observed frequency is periodically updated so that a given channel always retains the same velocity, not the same frequency.

16. L. Young, Synthesis Imaging Summer School, 19 June 2002 16 Doppler Tracking: Why should I care? The bandpass shape is really a function of the observed frequency, not the observed velocity. Extremely accurate bandpass calibration will require that a given channel in the calibrator data corresponds to the same frequency as the channel in the source data. This is a bigger deal at high frequencies than at low frequencies because a fixed velocity difference is a larger fraction of the total observed bandwidth. Projects desiring very high spectral dynamic range or projects at high frequencies (VLA’s 1 cm and 7 mm bands) will have to pay attention to whether Doppler tracking is on or off.

17. L. Young, Synthesis Imaging Summer School, 19 June 2002 17 Correlator Setups: Bandwidth Coverage and Velocity Resolution Your science will dictate how much frequency space you need to cover and how finely you need to cover it. For example, if you are studying a galaxy whose HI line is 200 km/s wide, you will want > 1 MHz bandwidth. If you are looking for departures from circular motion in that galaxy you may want spectral resolution of 5 km/s or better.

18. L. Young, Synthesis Imaging Summer School, 19 June 2002 18 Example: available bandwidths and spectral resolutions at the VLA

19. L. Young, Synthesis Imaging Summer School, 19 June 2002 19 Getting spectral information from the lags (XF correlator) Consider monochromatic signals of period T from two antennas. Signal 2 shifted by lag  multiply; integrate. Signal 2 shifted by lag  multiply; integrate. Signal 2 shifted by lag  T/2; multiply; integrate. X ij (  X ij (  T/4)X ij (  = T/2) …… Fourier transform of lag spectrum X ij (  ) is source spectrum V ij ( 

20. L. Young, Synthesis Imaging Summer School, 19 June 2002 20 Constraints on bandwidth and spectral resolution (Nyquist sampling theorem) to recover total bandwidth  BW, must use lag spacing no larger than   BW. Larger bandwidth requires smaller lags. takes a longer time to distinguish between two very closely spaced frequencies. To achieve frequency resolution  CH, must sample the lag spectrum out to N  CH. Finer frequency resolution requires more lags (or fatter lags). Finite correlator power leads to tradeoffs between total bandwidth covered and frequency resolution (and number of polarization products).

21. L. Young, Synthesis Imaging Summer School, 19 June 2002 21 The Gibbs Phenomenon: spectral ringing Consider a source with strong continuum and a strong, narrow spectral line. Remember that the source spectrum is the Fourier transform of the lag spectrum; Also, synthesizing very sharp edges requires an infinite number of Fourier components. We don’t have that many.

22. L. Young, Synthesis Imaging Summer School, 19 June 2002 22 The Gibbs Phenomenon and what to do about it The result of trying to synthesize sharp edges with a finite number of lags. Fixes for Ringing: 1. Hanning smoothing (online or offline); spectral resolution gets worse by x2 2. Ignore it. If the continuum is not strong and your line is not strong/sharp, it may not be an issue.

23. L. Young, Synthesis Imaging Summer School, 19 June 2002 23 Correlator Setups: Bandwidth Coverage and Velocity Resolution Check the Observational Status Summary (or equivalent document) carefully! You may be able to do two spectral lines simultaneously, or multiple bandwidths or resolutions. Always count on not being able to use the channels at the ends of the band. (At VLA: 1/8 of the channels at each end) The quality of your final image will partly depend on how well you can remove the continuum emission. Use enough bandwidth to get plenty of line-free channels.

24. L. Young, Synthesis Imaging Summer School, 19 June 2002 24 Chapters by Rupen and Westpfahl in SIRA II Lectures by Hibbard and Peck, this conference VLA Observational Status Summary: http://www.aoc.nrao.edu/vla/obstatus/vlas/vlas.html http://www.aoc.nrao.edu/vla/obstatus/splg1 http://www.aoc.nrao.edu/vla/html/highfreq/hfsched.html The “3MHz ripple”: Carilli (1991), VLA Test Memo #158 (see AOC library) Resources for Spectral Line Observers